|Publication number||US5136696 A|
|Application number||US 07/211,977|
|Publication date||Aug 4, 1992|
|Filing date||Jun 27, 1988|
|Priority date||Jun 27, 1988|
|Publication number||07211977, 211977, US 5136696 A, US 5136696A, US-A-5136696, US5136696 A, US5136696A|
|Inventors||Robert F. Beckwith, Neil J. Johnson, Suren Irukulla, Steven Schwartz, Nihar Mohapatra|
|Original Assignee||Prime Computer, Inc.|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (22), Non-Patent Citations (12), Referenced by (172), Classifications (9), Legal Events (14)|
|External Links: USPTO, USPTO Assignment, Espacenet|
This invention relates to high speed, pipelined digital processors and, more particularly, to pipelined central processors that utilize an instruction cache memory for storage of both program instructions and a multicycle instruction interpreter, and a cache memory for predicting multicycle instructions in the instruction flow.
One form of architecture used in high speed computers is pipelining. A pipelined central processor is organized as a series of stages where each stage performs a dedicated function, or task, much like a job station on a factory assembly line. While pipelining does not decrease the time required to execute an instruction, it does allow multiple instructions in various phases of execution to be processed simultaneously. For an n stage pipeline, where one instruction enters the pipeline every cycle and a different instruction exits from the pipeline on every cycle, instructions appear to be executed at a rate equal to the cycle time. The actual execution time is n times the cycle time, as long as there are no dependencies between instructions in the program being executed.
Quite frequently, however, execution of one instruction requires data generated by another instruction. Execution of the instruction requiring the data must be held up until the instruction generating the data is completed. Typically, dependencies arise in (1) instruction sequencing, for example, a conditional branch; (2) operand address formation, for example, loading a register used in forming an address; and (3) execute data, for example, calculating data that is used in a later operation. The delay associated with each of these dependencies is dependent upon type of dependency, the length of the pipeline and the spacing between the two instructions in the program. The delay is increased for longer pipelines and/or close spacing between two instructions with dependencies. Several techniques have been developed to minimize the delay, including branch predictors and bypass paths which short circuit the normal pipeline data flow.
Many current computer architectures contain instructions which require several processor cycles to execute. This type of architecture is commonly known as a "complex instruction set computer" (CISC). The term CISC arose as the popularity of another type of architecture, "reduced instruction set computers" (RISCs), grew. The predominance of single cycle instructions and the lack of complex pipeline interlock detection hardware are characteristic of the RISC machines.
Program execution in a data processor usually includes fetching programmed instructions from memory and then performing the operation indicated by that instruction. An instruction decoder portion of the processor hardware is dedicated to transforming the ones and zeros in an instruction into signals used to control operations performed by other parts of the processor.
To execute single-cycle instructions, the instruction decoder can be a simple lookup table addressed by the instruction. The table width is determined by the required number of control signals. The table can be implemented using logic gates or memory such as RAM or ROM.
The execution of multicycle instructions requires a control sequencer which can be implemented using either logic gates or memory. Computers with only a few multicycle instructions may utilize the logic approach. As the number of multicycle instructions increases, the memory-based design is usually chosen, since the logic approach becomes very cumbersome. The memory-based design is usually called microcode. Most current computer designs, with the exception of RISC machines, rely heavily upon microcode and microprograms to provide the desired processor functionality. The incorporation of microcode into a machine design requires a memory in which to store the microcode, usually called the control store, and a sequencer to provide flow control for the microcode.
Pipelined computers typically include at least an instruction fetch and decode stage and an execute stage. Microcode has been incorporated into the execute stage for handling of multicycle instructions. When a multicycle instruction is decoded, the microcode in the execute stage sequences through the steps of the complex instruction, and the remainder of the pipeline is placed on hold. Disadvantages of the microcode approach include the additional hardware required for implementing the microcode and somewhat slower operation due to memory latency involved in execution of the microcode.
Other approaches to the execution of multicycle instructions have included the storage of millicode instructions in memory, as described by D. S. Coutant et al in Hewlett Packard Journal, Jan. 1986, pp. 4-19. Still another approach described by A Bandyopadhyay et al in "Combining Both Micro-code And Hardwired Control in RISC," Computer Architecture News, Sep. 1987, pp. 11-15, involves the incorporation of a bit into each instruction that indicates whether the instruction is single cycle or multicycle. When the bit indicates a multicycle instruction, microcode is called. The disadvantage of both these approaches is a delay in operation of one or two machine cycles, since the instruction is not identified as a multicycle instruction until it has been decoded. Thus, the subsequent cycle must wait until the preceding instruction is decoded.
The use of a branch cache prediction mechanism for providing more efficient pipelined processor operation is described in U.S. Pat. No. 4,777,594 issued Oct. 11, 1988 and assigned to the assignee of the present application. Without a prediction mechanism, a pipelined processor may function slowly on a branch instruction. A branch instruction requires branching to an instruction out of the normal sequence. However, earlier pipeline stages have begun processing the next instruction in the normal sequence. When this occurs, flushing of the pipeline and restarting is required, and processing is delayed. The branch cache memory is associated with instruction fetching and predicts the next address after a branch instruction based on past operation. Thus, when a branch instruction is encountered in the program, the branch cache predicts the target address of the branch instruction, and the next instruction address in the normal sequence is replaced with the branch target address. As a result, pipeline operation proceeds without interruption.
It is a general object of the present invention to provide improved digital processing apparatus.
It is another object of the present invention to provide a pipelined central processor capable of high speed instruction processing.
It is a further object of the present invention to provide a high performance, pipelined central processor capable of executing both single-cycle instructions and multicycle instructions.
It is yet another object of the present invention to provide a high performance, pipelined central processor that is simple in construction and low in cost.
It is still another object of the present invention to provide a high performance central processor that utilizes a cache memory for predicting multicycle instructions and for calling an instruction interpreter located in an instruction cache memory.
According to the present invention, these and other objects and advantages are achieved in a high-performance pipelined central processor capable of efficiently executing both single-cycle instructions and multicycle instructions. The pipelined processor includes at least an instruction fetch stage, an instruction decode stage, an address formation stage, an operand stage, an execute stage and a write stage. The instruction fetch stage includes an instruction cache memory and a prediction cache memory that are commonly addressed. Instructions of the program being executed are held in the instruction cache memory. An instruction interpreter for executing multicycle instructions is also located in the instruction cache memory.
The instruction interpreter includes a sequence of microinstructions for executing each multicycle instruction. The entry address for the interpreter is located in the prediction cache memory. When a complex instruction occurs in the program, the prediction cache memory makes an interpreter call prediction based on the instruction address and points to the interpreter entry address. The program return address is pushed onto a program counter stack, and the interpreter entry address is loaded into the program counter. The microinstructions of the interpreter are fetched sequentially from the instruction cache memory by the program counter. When the interpreter is completed, the prediction cache memory makes a return prediction, and the program return address is transferred from the stack to the program counter. By use of the prediction cache memory, calls to the interpreter and returns from the interpreter are performed without loss of processing cycles.
According to the invention, there is provided digital processing apparatus for executing stored program instructions including single-cycle instructions and multicycle instructions during multiple processing cycles. The processing apparatus comprises instruction memory means for providing the program instructions in response to program addresses and for providing microinstructions of a multicycle instruction interpreter in response to microinstruction addresses, prediction means responsive to different ones of the program addresses for making predictions of multicycle instructions and predictions of single cycle instructions, control means for providing consecutive program addresses to the instruction memory means for accessing the program instructions in response to predictions of single cycle instructions and for providing consecutive microinstruction addresses to the instruction memory means in response to each prediction of a multicycle instruction until completion of the multicycle instruction interpreter, means for executing the program instructions and the microinstructions and means for validating the predictions and for updating the prediction means when the predictions are incorrect.
In a preferred embodiment, the predictions include interpreter call predictions and interpreter return predictions, and the prediction means includes a cache memory for storage and readout of each of the call predictions and an associated entry address at the program address for which the call prediction is made and for storage and readout of each of the return predictions at the microinstruction address for which the return prediction is made. The prediction cache memory can further include means for storage and readout of branch instruction predictions. Each branch instruction prediction is stored with a branch target address at the program address for which the branch prediction is made. When a branch prediction is made, the target address is loaded into the program counter for addressing of the instruction cache memory.
The above-described control means preferably includes program counter means for holding a current address for accessing the instruction memory means on each processing cycle and for incrementing the current address on consecutive processing cycles, stack memory means coupled to the program counter means, control logic means responsive to each of the call predictions for loading a return program address from the program counter means into the stack memory means and selector means responsive to each of the call predictions for loading the entry address into the program counter means and responsive to each of the return predictions for transferring the return program address from the stack memory means to the program counter means.
For a better understanding of the present invention together with other and further objects, advantages and capabilities thereof, reference is made to the accompanying drawings which are incorporated herein by reference and in which:
FIG. 1 is a simplified block diagram that illustrates the pipelined processor architecture of the present invention;
FIG. 2 is a simplified block diagram that illustrates the instruction fetching unit of the pipelined processor;
FIGS. 3a, 3b and 3c are a block diagram that illustrates program sequencing of the pipelined processor;
FIG. 4 is a timing diagram that illustrates an example of the operation of the instruction fetching unit of FIG. 2; and
FIGS. 5a and 5b are a block diagram that illustrates the portion of the pipelined processor downstream from the instruction fetching unit.
A pipelined processor incorporating the present invention is designed as an eight-stage pipeline. Execution of each machine instruction is divided into a minimum of eight steps which are performed in eight consecutive clock cycles. When an instruction enters stage two of the pipeline, the following instruction enters stage one. In an ideal situation, the staging of instructions continues until the pipeline is full, which permits eight instructions to be processed concurrently. When the pipeline is full, an instruction completes execution every clock cycle. In actual operation, cache misses, dependencies, incorrect predictions and other exceptional conditions cause the pipeline operation to be interrupted, so that the average instruction time is about three clock cycles. In a preferred embodiment, the clock cycle is eighteen nanoseconds. It will be understood that different clock cycle times can be utilized, depending on the hardware speed and the processing requirements.
A simplified block diagram of the eight stage pipeline processor is shown in FIG. 1. A first pipeline stage 10 contains an instruction cache memory, a branch cache memory and the hardware to address the cache memories, as described hereinafter. The instruction cache stores the instructions for the program being executed, and also stores an interpreter for multicycle instructions. The branch cache memory is a mechanism for predicting branch instructions and multicycle instructions, based on the instruction address and the past history of that instruction address. In the present embodiment, the instruction cache is a two-way set associative cache. A second pipeline stage 12 performs instruction cache set selection and generates a hit/miss signal. If a single set instruction cache is utilized, the second pipeline stage 12 can be omitted from the processor. In a third pipeline stage 14, the instruction is decoded to provide control signals and addresses for the remainder of the processor.
A fourth pipeline stage 16 determines the effective address of the operand by performing up to a three way addition of a base register plus an index register plus a displacement. A fifth pipeline stage 18 is an operand cache, and a sixth pipeline stage 20 performs an operand cache set selection. The operand cache, which stores the operands for the program, includes multiple sets and is virtually addressed. The operand cache set select stage 20 performs operand cache hit/miss determinations, set selection and data alignment. If a single set operand cache is utilized, the sixth pipeline stage 20 can be omitted from the processor. An execute stage 22 performs the operation specified by the instruction, and a write stage 24 stores the results determined in the execute stage 22.
A more detailed block diagram of the instruction portion of the pipelined processor is shown in FIG. 2. In pipelined processor architecture, each stage of the pipeline typically includes combinatorial logic followed by a latch, or register, at its output for holding the results. The outputs of each register supply inputs to the following stage. Referring now to FIG. 2, a program counter register 40 is used to fetch instructions from an instruction cache memory 42. The outputs of program counter register 40 are coupled by a bus 44 to the address inputs of instruction cache 42. The bus 44 also couples the outputs of the program counter register 40 to address inputs of a branch cache memory 46 and to inputs of an adder 48.
The program counter register 40 receives inputs on a bus 50 from a data selector 52. Data selector 52 loads the program counter register 40 with the address of the instruction to be accessed in instruction cache 42. The program counter register 40 input is selected from one of four sources. The outputs of adder 48 are coupled by a bus 54 to a first set of inputs of data selector 52. When this input is selected, program counter register 40 is incremented by a predetermined number, typically two, after each instruction. Program counter register 40 and adder 48 together function as a program counter for sequencing through addresses of program instructions. Bus 54 also couples the output of adder 48 to a stack memory 56. The output of stack 56 is coupled by a bus 58 to a second set of inputs of data selector 52. When the stack 56 output is selected by data selector 52, the top entry on stack 56 is loaded into program counter register 40 as the address of the next instruction. The output of branch cache 46 is coupled by a bus 60 to a third set of inputs of data selector 52. When the output of branch cache 46 is selected, a predicted instruction address other than the next address is predicted, as described in detail hereinafter. A fourth set of inputs to data selector 52 is a trap vector that is utilized in the event of certain faults during operation.
A program sequencer 61 receives prediction types from the output of branch cache 46 and supplies control signals to data selector 52 and to stack 56. The program sequencer 61 controls storage of return addresses in stack 56 and selection of the appropriate instruction address by data selector 52.
The output of the instruction cache 42 is held in an instruction buffer 62 that has its outputs coupled to to an instruction cache set select and align unit 64 (FIG. 3). The output of unit 64 is coupled through an instruction register 66 to an instruction decoder 68. The set select and align unit 64 determines cache hit or cache miss and selects one output of the multiset instruction cache memory. Instructions to be executed are passed in sixteen-bit parcels from the instruction cache 42 to the instruction decoder 68 through the set select and align unit 64. The instruction decoder 68 assembles the instruction parcels into complete instructions and then extracts all the information required to successfully complete execution of the instruction. Most of the information is contained in RAM tables within the instruction decoder 68. The outputs of instruction decoder 68 are coupled through a decoder register 70 to the remainder of the pipelined processor.
Instruction sequencing has been described above in connection with program counter register 40 used for addressing of the instruction cache 42 and the branch cache 46. The instruction sequencer for the pipelined processor of the present invention includes one program counter register per stage of the pipelined processor. As shown in FIG. 3, the program counter register 40 is associated with the first pipeline stage 10. Stages 2 through 8 of the pipelined processor contain program counter registers 82, 84, 86, 88, 90, 92 and 94, respectively. Program counter register 94 associated with write stage 24 is the real program counter. There is also provided a program counter register 96 that is used to back up program counter register 94. The contents of program counter register 40 are cycled through the stages of the pipelined processor in phase with the corresponding instructions.
The processor of the present invention maintains two physical stacks 56 and 98 that are used in implementing pipeline flushing. Stack 56 is associated with the instruction fetch stage 10 and is used to push or pop return addresses to or from program counter register 40. Stack 98 is associated with the write stage 24 and is used to push or pop return addresses to or from program counter register 94. Stacks 56 and 98 are coupled together by a bus 102. When a pipe flush condition occurs, the entire contents of the write stage stack 98 are transferred via bus 102 into the instruction prefetch stage stack 56. Hence, the predicted stack operations are undone and the integrity of the stack is maintained.
The branch cache 46 is used to predict both branch instructions and multicycle, or complex, instructions. Thus, although the cache memory 46 is called a branch cache, it is not limited to predicting branch instructions. The branch cache 46 is also capable of call predictions and return predictions as described in detail hereinafter. The use of a branch cache for prediction of branch instructions is disclosed in the aforementioned U.S. Pat. No. 4,777,594. When the cache 46 predicts a multicycle instruction, a call is made to an instruction interpreter stored in the instruction cache 42. The instruction interpreter is a series of microinstructions utilized for executing a multicycle instruction. The instruction interpreter stored in the instruction cache 42 is analogous in function to the microcode used on prior art processor designs. However, the instruction interpreter utilized in the processor of the present invention provides improved performance relative to architectures where a microsequencer is part of the execution stage of the pipeline.
The instruction cache 42 stores both the normal program instruction sequence and the instruction interpreter. This is possible because the two sets of instructions are accessed at different times. When a normal program instruction sequence is being accessed, the interpreter is not required. When a multicycle instruction is predicted by the branch cache 46, the normal instruction flow is delayed while the steps of the complex instruction are executed. Therefore, the available bandwidth of the entire pipeline is more efficiently utilized.
The branch cache 46 is addressed by the program register 40 on every processor cycle. Each branch cache location contains two bits which indicate prediction type. Possible prediction types are (1) normal prediction, (2) branch prediction, (3) interpreter call prediction and (4) interpreter return prediction. In the case of a branch prediction, the branch cache 46 stores a sixteen-bit branch target address. In the case of an interpreter call prediction, the branch cache 46 stores a sixteen bit interpreter entry address. No data is stored with an interpreter return prediction or a normal prediction. Sixteen bits are adequate since the branch target address or the interpreter entry address is normally located in the same memory segment as the calling instruction. If this condition is not met, the branch cache output can be expanded to include more than sixteen bits.
In an important feature of the invention, the predictions made by branch cache 46 are based on instruction addresses rather than the instructions themselves. Predictions are stored in the branch cache 46 based on past operation of the system. Each prediction is stored in the branch cache 46 at the same address as the corresponding instruction is stored in the instruction cache 42. The branch cache 46 predicts that certain program addresses contain multicyle instructions, branch instructions, etc. The ability to make predictions from instruction addresses provides a considerable time saving and reduces processing time. With the predictions from branch cache 46, entry to the instruction interpreter, return from the instruction interpreter and branches can be performed on the processor cycle immediately following the prediction. Without the branch cache 46 for making predictions, one or more processing cycles would be lost, since it would be necessary to wait until the instruction was decoded before accessing the interpreter or the branch target.
In the case of a normal prediction, operation continues in a normal program mode, and the program counter register 40 is incremented by the adder 48. In the case of a branch prediction, the branch target address is loaded into the program counter register 40 through the data selector 52, and the branch target instruction is addressed on the processor cycle immediately after the instruction for which the branch was predicted.
In the case of a multicycle instruction, the branch cache 46 stores a call prediction and an entry address of the instruction interpreter. The entry address may be different for each different type of multicycle instruction. The interpreter entry address is loaded into program counter register 40 through data selector 52. The entry address of the instruction interpreter is placed in program counter register 40 on the processor cycle immediately after the instruction that was predicted to be a multicycle instruction. The processor then enters an interpreter mode, and microinstructions of the interpreter are accessed consecutively. The microinstructions of the instruction interpreter are sequenced in the same manner as a normal instruction sequence. The program counter register 40 is incremented by the adder 48 after each microinstruction. When the interpreter entry address is loaded from the branch cache 46 into program counter register 40, the return address of the next normal instruction is pushed onto stack 56. The stack pointer is incremented, and an interpreter mode bit is set. As described hereinafter, the interpreter mode bit is derived from the branch cache 46 and indicates whether the instruction is a normal program instruction or an interpreter microinstruction. The return address is the current address in program counter register 40 incremented by adder 48 and appears at the output of adder 48 on bus 54. Normal instruction sequencing is delayed during addressing of the instruction interpreter.
During sequencing through the microinstructions of the interpreter, the branch cache 46 is being addressed. Branch cache 46 predicts completion of the interpreter by means of a return prediction. The return prediction causes the return address at the top of stack 56 to be loaded into program counter register 40 through data selector 52. The stack pointer is decremented, and the interpreter mode bit is reset if the stack pointer is zero. As a result, the program returns to normal instruction flow until the next branch instruction or multicycle instruction is encountered. The return to normal instruction flow occurs on the processing cycle immediately after the last microinstruction of the interpreter.
The instruction interpreter that is stored in instruction cache 42 can include subroutines. Subroutine calls within the instruction interpreter are handled in the same manner as calls to the interpreter. When a subroutine is called during execution of the interpreter, the subroutine call is predicted by the branch cache 46 as a call prediction. The branch cache 46 stores a subroutine entry address at the location of the subroutine call prediction. The return address of the next microinstruction is pushed onto the top of the stack 56, the stack pointer is incremented, and the subroutine entry address is loaded from the branch cache 46 into the program counter register 40 through data selector 52. When the subroutine is completed, branch cache 46 makes a subroutine return prediction, the stack pointer is decremented, and the subroutine return address is transferred from stack 56 through data selector 52 to program counter register 40. The instruction interpreter is then completed, and an interpreter return prediction is made. At this time, the interpreter return address is transferred from the stack 56 to the program counter register 40 for return to the normal program flow. It will be understood that the interpreter can include nested subroutines that are limited in number only by the capacity of the stack to store the return addresses.
In order to keep the normal program mode separate from the interpreter mode, the branch cache 46 is divided into two sections: a normal program section and an interpreter section. The two sections are addressed by the interpreter mode bit, which is derived from the branch cache output on previous instructions. During normal program flow, the interpreter mode bit is reset. When an interpreter call is made due to the branch cache 46 predicting a multicycle instruction, the interpreter mode bit is set. The interpreter mode bit remains set during subsequent microinstructions of the interpreter and is reset upon a return from the interpreter. The interpreter mode bit is one of the address bits supplied to the branch cache 46. The normal program section of the branch cache 46 (addressed by the reset state of the interpreter mode bit) contains program branch predictions and interpreter call predictions. The interpreter section of the branch cache 46 (addressed by the set state of the interpreter mode bit) contains the following types of predictions.
1. Interpreter branch predictions (branches within the interpreter).
2. Interpreter subroutine calls.
3. Interpreter returns (either return to normal program mode or return from interpreter subroutines).
While the program predictions and interpreter predictions are maintained in separate sections of the branch cache 46, the outputs of the branch cache 46 are differentiated only as to normal prediction, branch prediction, interpreter call prediction and interpreter return prediction.
In addition to the above-described operations that occur when a prediction is made, the following additional operations can be performed on the program counter register 40.
1. The program counter register 40 can be held in the same state for two or more processing cycles.
2. In a flush operation, the program register 40 is loaded with the contents of the write stage program register as described hereinafter.
3. In a cache miss operation, the program register 40 is loaded with the address values in the second stage program counter register 82.
In a preferred embodiment, the instruction cache 42 includes a virtually addressed two-way set associative cache with 16K bytes per set and a two way set associative segment table look-aside buffer with 512 entries per set. The instruction cache keeps a local copy of the most recently-used instructions in high speed RAM. The branch cache 46 is a two way set associative cache with 1K entries per set. Of the 1K entries, 512 are reserved for program branch and interpreter call instructions, and the other 512 are reserved for interpreter branch, interpreter subroutine call and interpreter return instructions, as described above.
Operation of the instruction fetching unit shown in FIG. 2 and described hereinabove is now illustrated with reference to FIG. 4, which indicates the states of program counter register 40, instruction buffer 62, branch cache 46, stack 56, adder 48 and data selector 52 for consecutive processing cycles T1, T2, T3, etc. During processing cycle T1, instruction buffer 62 contains a single-cycle instruction S0 and program counter register 40 contains program address AS1 of the next single-cycle instruction. During the next processing cycle T2, the single-cycle instruction S1 is loaded into instruction buffer 62, and the next program address AS2 is held in program counter register 40. During processing cycles T1 and T2, normal predictions N are made by branch cache 46, and the adder 48 increments the program counter register 40 by two counts on each processing cycle.
During processing cycle T3, single-cycle instruction S2 is moved into instruction buffer 62, and the next program address AM1 is moved into program counter register 40. Address AM1 corresponds to a multicycle instruction M1. The branch cache 46 makes an interpreter call prediction C and provides an interpreter entry address AI1. As a result of the call prediction, the data selector 52 selects inputs from branch cache 46 on processing cycle T4. The entry address AI1 is loaded from branch cache 46 through data selector 52 to program counter register 40, and a return address AS3 (which equals address AM1 +2) is pushed onto stack 56. During processing cycles T5 to T7, the interpreter microinstructions I1, I2 and I3 are addressed consecutively by means of the adder 48 incrementing the program counter register 40. During processing cycle T6, branch cache 46 makes a return prediction R. As a result of the return prediction, the data selector 52 selects inputs from stack 56. On subsequent processing cycle T7, the return address AS3 at the top of stack 56 is transferred to program counter register 40, and normal program sequencing continues.
During processing cycle T8, the program address AB1 in program counter register 40 is a branch instruction B1, and the branch cache 46 provides a branch prediction B and a branch target address AT1. As a result of the branch prediction, the data selector 52 selects inputs from branch cache 46, and on subsequent processing cycle T9, the branch target address AT1 is transferred from branch cache 46 to program counter register and normal program sequencing continues.
During processing cycle T11, an address AM2 corresponding to a multicycle instruction M2 is loaded into program counter register 40. In this case, the branch cache 46 incorrectly makes a normal prediction N rather than an interpreter call prediction C. The handling of incorrect predictions is described hereinafter.
Thus far, it has been assumed that the branch cache 46 contains valid information for making predictions of branch instructions and interpreter call predictions and return predictions. The branch cache is a means for predicting or guessing the occurrence of branch instructions and multicycle instructions in a program, based on past program operation. The predictions reduce processing time when they are correct. In each instance, the prediction must be validated by subsequent comparison with the actual program instruction after the instruction is accessed. When an incorrect prediction is made, the branch cache 46 must be updated to reflect the correct information, and the pipeline must be flushed so that the incorrectly predicted operation can be performed correctly. The updated information increases the probability that correct predictions will be made during future operations.
During the first cycle after a program or a program segment is called from memory, no predictions are contained in the branch cache 46. In this situation, incorrect predictions will be made for every branch instruction and multicycle instruction. Updates of the branch cache 46 are performed in each case of an incorrect prediction. Then, on subsequent program cycles, predictions are correctly made.
There are three possible mistakes that can be made by the branch cache 46: (1) a bad prediction; (2) an effective address formation (EAF) gaffe; or (3) an instruction gaffe. A bad prediction occurs when the branch cache 46 incorrectly predicts the result of a conditional branch instruction. Bad predictions can only be validated in the execution unit of the pipelined processor. In the execute unit, the condition codes are tested to determine if the branch cache 46 predicted the correct operation. EAF gaffes are detected by comparing the branch target address generated by the effective address formation unit 16 with the branch target address predicted by the branch cache 46. EAF gaffe detection is performed in the stage immediately following the effective address formation stage 16. Instruction gaffes are detected by comparing the instruction type from the decode stage 14 with the instruction type predicted by the branch cache 46. For example, if the branch cache 46 predicts a branch on a nonbranch instruction, then an instruction gaffe is detected. The instruction gaffe detection is determined in the stage immediately following the instruction decode stage 14. The entry address of the interpreter is available at the output of the decode stage 14. The predicted entry address and the decoded entry address are compared to determine validity.
The program counter registers 40, 82, 84, 86, 88, 90, 92 and 94 that are associated with each stage of the pipeline describe the state of the instruction within each stage. Each program counter register contains a address field and a status field for each instruction in the pipeline. As an instruction moves down the pipeline, its status and address fields move with it.
The address field of the program counter register is used for different purposes in different stages of the pipeline. In the first stage, program counter register 40 addresses the instruction cache 42. In stages two through four of the pipeline, the program counter register address field is used for gaffe detection. Stages five through seven use the address field to send branch targets to program register counter 94. In the write stage 24, the address field is the real program counter which can be incremented by the instruction length code, loaded with branch/call targets, pushed/popped and loaded with trap vectors. In all stages, the program counter register represents the next address of each corresponding instruction in the pipeline.
The status field bits carried through the program counter registers are described below.
1. The alignment bit indicates the alignment of the associated instruction. The alignment bit is equal to zero when the instruction ends on the 32-bit boundary (aligned and otherwise is set to one.
2. The prediction bit indicates that the associated instruction has caused a branch cache prediction.
3. The least-recently used bit is used in the branch cache update/invalidate algorithm.
4. The history bit is used in the branch cache update/invalidate algorithm.
5. The trap bits cause an instruction to be trapped, either when it reaches the write stage or after it is completed in the write stage.
6. The serialization bit causes the pipeline to be flushed after the tagged instruction completes in the write stage.
7. The nullification bit causes the tagged instruction to be flushed from the pipeline and retried when the tagged instruction reaches the write stage.
8. The valid bit flags a stage as containing an instruction. When the bit is clear, it means that the stage is empty. Empty stages occur after each pipeline flush.
9. The gaffe bit is set by the gaffe hardware as described hereinafter. The gaffe bit causes the branch cache 46 to be updated when the tagged instruction reaches the write stage 24 and causes the tagged instruction to be serialized so that the pipeline is flushed after the instruction is completed.
10. The interpreter mode bit is used selectively put any stage of the pipeline in the interpreter mode. The interpreter mode bit comes from the branch cache 46.
11. The instruction length code (ILC) is associated with every instruction in the pipeline. Its value is used to increment program counter register 94 when an instruction is completed in the write stage 24.
12. The two-bit operation code tags each instruction as being a branch, call, return or other instruction. The operation code is set by the branch cache 46 and is used to validate branch cache predictions (instruction gaffe and conditional branch verification). In the EAF stage 16 of the pipeline, the operation code is used for instruction gaffe detection. After EAF stage 16, the bits of the operation code are used to pass the operation code from the decode stage 14 down the pipeline.
13. The set select bit identifies the set to invalidate on a branch cache gaffe or bad prediction.
Referring again to FIG. 3, the output of decoder register 70 is connected to one set of inputs of a comparator 80. When a call prediction or a branch prediction is made by the branch cache 46, the entry or target address is loaded into the address field and the prediction type is loaded into the operation code bits of the status field of program counter register 40. The prediction and the entry or target address are then sequenced through program counter registers 82 and 84 to a second set of inputs of comparator 80. Passing the branch cache 46 outputs program counter registers 40, 82 and 84 insures that the inputs are provided to comparator 80 simultaneously. Instruction gaffe detection is performed by comparing the operation code (instruction type) from instruction decoder 68 with the operation code (prediction type) from program counter register 84. When the instruction type and the prediction type agree, the branch cache prediction was correct, and operation can continue. When these values do not agree, the branch cache 46 made an erroneous prediction, and the comparator 80 supplies an instruction gaffe signal which sets the gaffe bit in program counter register 86. When an interpreter call prediction is made by the branch cache 46, the comparator 80 also compares the entry address predicted by branch cache 46 with the entry address supplied from decoder 68. When these addresses do not agree, an instruction gaffe signal is provided.
The target address of a branch instruction is not available for validation until the output of the effective address formation stage 16. The output of decoder register 70 is coupled to effective address formation unit 72, and the effective address is stored in EAF register 74. The output of EAF register 74 is coupled to one set of inputs of a comparator 76. The output of stage four program counter register 86 is coupled to the other set of inputs of comparator 76. The program counter registers 40, 82, 84 and 86 insure that the branch target address prediction is supplied to the comparator 76 at the same time as the effective address from EAF register 74. EAF gaffe detection is performed by comparing the branch address from the EAF register 74 with the address field of program counter register 86. When the two inputs to comparator 76 are equal, the branch cache 46 correctly predicted the target address of the branch instruction, and operation can continue. When the inputs to comparator 76 are not equal, an incorrect target address prediction was made, and an EAF gaffe signal is provided.
If a branch prediction, call prediction or return prediction by the branch cache 46 was incorrect, the corresponding instruction is invalidated, but execution of the instruction is completed. It is necessary to complete the instruction in order to determine the correct result for updating the branch cache 46. In the case of a conditional branch instruction, the condition is not determined until the execute stage 22.
The write stage 24 contains the real program counter register 94 and is responsible for advancing the CPU from its current state to its next state for each instruction executed. The write stage 24 stores all operand results, sets condition codes and loads the program counter register 94 with the next instruction address. When previous stages of the pipelined processor make an error, it is the responsibility of the write stage 24 to flush the pipeline. Flushing of the pipeline consists of invalidating all the instructions in the first seven stages of the pipeline and loading the contents of program counter register 94 into program counter register 40. The most common causes of pipeline flushes are bad branch cache predictions and gaffes.
Two types of pipeline flushing are utilized. In serialization, the instruction causing the flush is completed. In nullification, the instruction causing the flush is not completed. In the case of serialization, all stores and other operations associated with an instruction are completed before the next instruction is fetched from memory. After completing the instruction in the write stage, the flush is performed by clearing the valid bit in the first seven stages of the pipeline. Flushing also causes the contents of program counter register 94 to be loaded into program counter register 40. Serialization is used to reset the pipeline when the branch cache 46 has made an incorrect prediction. In parallel with serialization, the branch cache 46 is updated. When serialization is performed, all pending operations in the stack 56 are also flushed.
In the case of nullification of an instruction, no stores associated with an instruction are completed and program counter register 94 is not updated. Nullification is different from serialization in that all stages of the pipeline are flushed including the write stage 24. Hence, the instruction in the write stage is not allowed to be completed. Instructions can be tagged for nullification in any stage of the pipeline and when the tag instruction reaches the write stage 24, it is nullified.
The write stage 24 performs several different operations depending on the state of the processor. Continue is the normal operation where the program counter register 94 is incremented by the instruction length code (ILC) and instruction processing continues. If the branch cache 46 predicted an operation other than continue, then the pipeline is serialized (flushed) and the branch entry is invalidated/updated.
Idling occurs when no instructions are passing through the write stage 24 due to holes (empty stages) in the pipeline. When a hole passes through the write stage 24, the write stage idles. Stages are empty when their valid bit is reset. Hold is treated exactly like idle, except that it is due to a hold condition such as instruction buffer full.
Branching is performed by loading the program counter register 94 with the branch target address. Branch target addresses come from the EAF stage 16. Branch target addresses are passed from the EAF stage 16 to the write stage program counter register 94 through program counter registers 88, 90 and 92. If the branch cache 46 predicted an operation other than a branch, then the pipeline is serialized (flushed), and the branch cache entry is updated/invalidated.
Interpreter calls are performed by loading program counter register 94 with the entry address. Entry addresses come from the EAF stage 16. Entry addresses are passed from the EAF stage 16 to the program counter register 94 through program counter registers 88, 90 and 92. The old value of the program counter register 94 is incremented by the ILC and is pushed on to the stack 98. If the branch cache 46 predicted an operation other than an interpreter call, then the pipeline is flushed, and the branch cache entry is updated with the call prediction.
Traps cause a trap vector address to be loaded into the program counter register 94. The old program counter register 94 value is incremented by the ILC and is pushed onto the stack 98. Traps always cause the pipeline to be flushed.
Interpreter returns are performed by popping the return address from the top of the stack 98 and loading it into the program counter register 94. If the branch cache 46 predicted an operation other than an interpreter return, then the pipeline is serialized (flushed), and the branch cache entry is invalidated.
Branch cache updating/invalidating takes one processor cycle to perform and is performed in parallel with serialization (flushing) in the write stage 24. To perform a branch cache update, a tag address and a target address are required. The tag address is the address of the branch instruction. The target address is the target address of the branch instruction. The update is performed by addressing the branch cache 46 with the tag address and then loading both the tag address and the target address into the branch cache 46. Branch cache invalidating is similar, but only requires the tag address. The code for a normal prediction is written into the prediction-type field of the branch cache 46 during an invalidation.
Updating of the branch cache 46 with a call prediction is performed in a similar manner. The tag address is the address of the multicycle instruction. The target address is the entry address of the interpreter. Updating the branch cache 46 for a return prediction requires only the tag address of the return, since the return address is contained in the stack. The branch cache update is performed by the write stage 24. When an instruction completes execution in the write stage, the program counter register 94 contains the target address needed for the branch cache update. Updating also requires the prediction type from the operation code to be loaded into the specified branch cache address.
The downstream stages of the pipelined central processor are illustrated in block diagram form in FIG. 4. It will be understood that the instruction fetching and sequencing apparatus shown and described hereinabove can be utilized with different downstream pipeline configurations.
The instruction decoder 68 receives program instructions and microinstructions from the instruction fetching unit and assembles and decodes instructions on every cycle where possible. The decoder 68 provides the necessary controls to execute single-cycle instructions in one cycle and supports the microinstructions used to execute multicycle instructions. Included in the instruction decoder 68 are alignment hardware to extract instructions from a stream of evenly-addressed parcels from the instruction cache 42, logic to reduce the opcode into an address to instruction control RAMs, logic to reduce the opcode into control signals for the EAF stage 16, logic to reduce the opcode into three E-unit register addresses (one write register and two read registers), logic to reduce the opcode into two I-unit register addresses (base and index), conflict/bypass detection for the I-unit registers and an 8K by 59 instruction control RAM to provide control signals for the sequencer, the operand cache, the execute unit and to generate an interpreter entry address and an execute unit microsequencer entry address.
Interpreter calls are made by the instruction decoder 68 when a multicycle instruction is detected in the instruction flow. The instruction decoder 68 provides an entry address into the interpreter and saves information in the instruction word which may be required by the interpreter (register file addresses, displacement, EAF information etc.). The entry address is sent to the EAF stage 16 for branch cache validation and possible update. On an incorrect prediction, the branch cache 46 must be updated and the pipeline must be flushed as described above.
In the effective address formation stage 16, a three-way addition between the base register or program counter word, number fields, index register word number and the displacement field extracted from the instruction is performed by a three operand arithmetic logic unit 122. Registers required for effective address formation, base, index and general registers are maintained in a triple port I-unit register file 120 which has two read ports and one write port. Also included in the EAF stage 16 is address staging. Since an interpreter routine may address relative to the effective address formation of the calling instruction, that address must be saved. The EAF stage 16 contains a program counter which is used for program counter relative addressing. The program counter in the EAF stage 16 is actually a pair of counters, the normal instruction counter and an interpreter counter. When the interpreter is entered, the normal counter is looped back onto itself and does not increment during interpreter routines. When the control is returned from the interpreter, the normal counter is incremented and the interpreter counter is not. Thus, a return from the interpreter simply entails a switch of counters, whereas a call to the interpreter requires a load to the interpreter counter and a switch of the counters. The arithmetic logic unit 122 has three basic input legs: base, index and displacement. The base and index legs are sourced from the instruction in the I unit register file 120, except when program counter relative addressing is in effect. In this case, the program counter offset field is coupled to EAF unit 74 (see FIG. 3), and the base leg is sourced from the program counter offset field. The displacement leg is sourced from the instruction decoder 68. The outputs of the arithmetic logic unit 122 are held in an EAF register 124.
The operand cache stage 18 includes an operand data cache and operand STLB 130. The operand data cache performs cache reads and writes, and the operand STLB performs virtual to physical address translation for cache operands. A store queue 132 buffers cache writes from the write stage 24 to the operand cache. The operand cache data is held in an operand buffer 134. The operand set select stage 20 includes operand format and alignment logic 140 which performs set selection and alignment on data from the operand stage. The results are held in an operand register 142.
The execute unit 22 performs the arithmetic and logical functions necessary to execute all instructions. It includes dedicated hardware for executing integer, floating point and decimal and character string instructions. The execute unit includes an E unit register file 150, a sequencer 152 and an execute unit control store 154. The arithmetic units are divided into an integer ALU 156, a floating point adder 158, a floating point multiplier 160, a divide unit 162 and a character and decimal unit 164. The outputs of these units are routed through a multiplexer 166 to an execute unit register 170. Most single-cycle and interpreter microinstructions are executed in one cycle. For these instructions, the sequencer 152 is used as a decoder. Most floating point instructions require multiple microinstructions, and the sequencer 152 is used to sequence through them. As described above, most multicycle instructions are handled by the interpreter in the instruction cache 42. However, it was found that floating point instructions may not operate efficiently when the interpreter described above is utilized, due to heavy data dependencies. Accordingly, floating point instructions are executed by the sequencer 152 and the E-unit control store 154 in the execute unit. Other instruction types reach the execute unit either as single cycle instructions or as interpreter microinstructions.
While there has been shown and described what is at present considered the preferred embodiments of the present invention, it will be obvious to those skilled in the art that various changes and modifications may be made therein without departing from the scope of the invention as defined by the appended claims.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US4179731 *||Apr 4, 1977||Dec 18, 1979||Tokyo Shibaura Electric Co., Ltd.||Microprogrammed control system|
|US4200927 *||Jan 3, 1978||Apr 29, 1980||International Business Machines Corporation||Multi-instruction stream branch processing mechanism|
|US4370711 *||Oct 21, 1980||Jan 25, 1983||Control Data Corporation||Branch predictor using random access memory|
|US4399507 *||Jun 30, 1981||Aug 16, 1983||Ibm Corporation||Instruction address stack in the data memory of an instruction-pipelined processor|
|US4454580 *||Sep 16, 1981||Jun 12, 1984||International Business Machines Corporation||Program call method and call instruction execution apparatus|
|US4477872 *||Jan 15, 1982||Oct 16, 1984||International Business Machines Corporation||Decode history table for conditional branch instructions|
|US4494187 *||Feb 22, 1982||Jan 15, 1985||Texas Instruments Incorporated||Microcomputer with high speed program memory|
|US4494188 *||Apr 5, 1982||Jan 15, 1985||Hitachi, Ltd.||Method of processing an operating system in a multi-processor system|
|US4498136 *||Dec 15, 1982||Feb 5, 1985||Ibm Corporation||Interrupt processor|
|US4583162 *||Jan 13, 1983||Apr 15, 1986||The Singer Company||Look ahead memory interface|
|US4679141 *||Apr 29, 1985||Jul 7, 1987||International Business Machines Corporation||Pageable branch history table|
|US4750112 *||Oct 23, 1986||Jun 7, 1988||Prime Computer, Inc.||Data processing apparatus and method employing instruction pipelining|
|US4760519 *||Sep 15, 1986||Jul 26, 1988||Prime Computer, Inc.||Data processing apparatus and method employing collision detection and prediction|
|US4764861 *||Feb 7, 1985||Aug 16, 1988||Nec Corporation||Instruction fpefetching device with prediction of a branch destination for each branch count instruction|
|US4777587 *||Aug 30, 1985||Oct 11, 1988||Advanced Micro Devices, Inc.||System for processing single-cycle branch instruction in a pipeline having relative, absolute, indirect and trap addresses|
|US4819234 *||May 1, 1987||Apr 4, 1989||Prime Computer, Inc.||Operating system debugger|
|US4847753 *||Jul 13, 1987||Jul 11, 1989||Mitsubishi Denki K.K.||Pipelined computer|
|US4853840 *||Jan 5, 1987||Aug 1, 1989||Nec Corporation||Instruction prefetching device including a circuit for checking prediction of a branch instruction before the instruction is executed|
|US4860197 *||Jul 31, 1987||Aug 22, 1989||Prime Computer, Inc.||Branch cache system with instruction boundary determination independent of parcel boundary|
|US4860199 *||Jul 31, 1987||Aug 22, 1989||Prime Computer, Inc.||Hashing indexer for branch cache|
|US4872109 *||Nov 2, 1987||Oct 3, 1989||Tandem Computers Incorporated||Enhanced CPU return address stack|
|US4942520 *||Jul 31, 1987||Jul 17, 1990||Prime Computer, Inc.||Method and apparatus for indexing, accessing and updating a memory|
|1||"4.2.2 Branch Target Cache", author and date unknown.|
|2||*||4.2.2 Branch Target Cache , author and date unknown.|
|3||A. Bandyopadhyay, "Combining both Micro-code & Hardwired Cont. in RISC", Comp. Arch. News, Sep. 1987, pp. 11-15.|
|4||*||A. Bandyopadhyay, Combining both Micro code & Hardwired Cont. in RISC , Comp. Arch. News, Sep. 1987, pp. 11 15.|
|5||D. S. Coutant et al, "Compilers for the New Generation of H-P Computers", H-P Journal, Jan. 1986, pp. 4-19.|
|6||*||D. S. Coutant et al, Compilers for the New Generation of H P Computers , H P Journal, Jan. 1986, pp. 4 19.|
|7||J. C. Circello, "Des. of the Edge 1 Supermini-comp.", VLSI Systems Des., May 1986, pp. 22-28.|
|8||*||J. C. Circello, Des. of the Edge 1 Supermini comp. , VLSI Systems Des., May 1986, pp. 22 28.|
|9||J. Cho et al, "The Memory Arch. & the Cache & Mem. Mgmt. Unit . . . ", U. of Cal. Report #UCB/CSD 86/289, p. 5.|
|10||*||J. Cho et al, The Memory Arch. & the Cache & Mem. Mgmt. Unit . . . , U. of Cal. Report UCB/CSD 86/289, p. 5.|
|11||R. D. Bernal, "Putting RISC Eff. to Work in CISC Arch.", VLSI Systems Des., Sep. 1987, pp. 46-51.|
|12||*||R. D. Bernal, Putting RISC Eff. to Work in CISC Arch. , VLSI Systems Des., Sep. 1987, pp. 46 51.|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US5222244 *||Dec 20, 1990||Jun 22, 1993||Intel Corporation||Method of modifying a microinstruction with operands specified by an instruction held in an alias register|
|US5230068 *||Feb 26, 1990||Jul 20, 1993||Nexgen Microsystems||Cache memory system for dynamically altering single cache memory line as either branch target entry or pre-fetch instruction queue based upon instruction sequence|
|US5257354 *||Jan 16, 1991||Oct 26, 1993||International Business Machines Corporation||System for monitoring and undoing execution of instructions beyond a serialization point upon occurrence of in-correct results|
|US5274792 *||Jan 21, 1993||Dec 28, 1993||Fujitsu Limited||Information processing apparatus with parallel instruction decoding|
|US5276822 *||Nov 14, 1990||Jan 4, 1994||Matsushita Electric Industrial Co., Ltd.||System with enhanced execution of address-conflicting instructions using immediate data latch for holding immediate data of a preceding instruction|
|US5283873 *||Jun 29, 1990||Feb 1, 1994||Digital Equipment Corporation||Next line prediction apparatus for a pipelined computed system|
|US5287522 *||Jun 29, 1990||Feb 15, 1994||Bull Hn Information Systems, Inc.||External procedure invocation apparatus utilizing internal branch vector interrupts and vector address generation, in a RISC chip|
|US5291586 *||Jul 17, 1992||Mar 1, 1994||International Business Machines Corporation||Hardware implementation of complex data transfer instructions|
|US5293499 *||Sep 21, 1990||Mar 8, 1994||Sun Microsystems, Inc.||Apparatus for executing a RISC store and RI instruction pair in two clock cycles|
|US5313634 *||Jul 28, 1992||May 17, 1994||International Business Machines Corporation||Computer system branch prediction of subroutine returns|
|US5361338 *||Nov 9, 1992||Nov 1, 1994||Mitsubishi Denki Kabushiki Kaisha||Pipelined system for decomposing instruction into two decoding parts and either concurrently generating two operands addresses of merging decomposing decoding codes based upon the second operand|
|US5381533 *||Mar 30, 1994||Jan 10, 1995||Intel Corporation||Dynamic flow instruction cache memory organized around trace segments independent of virtual address line|
|US5386519 *||Jul 23, 1992||Jan 31, 1995||Nec Corporation||Information processing apparatus incorporating buffer storing a plurality of branch target instructions for branch instructions and interrupt requests|
|US5396634 *||Sep 30, 1992||Mar 7, 1995||Intel Corporation||Method and apparatus for increasing the decoding speed of a microprocessor|
|US5404486 *||Jul 7, 1993||Apr 4, 1995||Sun Microsystems, Inc.||Processor having a stall cache and associated method for preventing instruction stream stalls during load and store instructions in a pipelined computer system|
|US5420990 *||Jun 17, 1993||May 30, 1995||Digital Equipment Corporation||Mechanism for enforcing the correct order of instruction execution|
|US5421020 *||Jan 8, 1993||May 30, 1995||International Business Machines Corporation||Counter register implementation for speculative execution of branch on count instructions|
|US5421022 *||Jun 17, 1993||May 30, 1995||Digital Equipment Corporation||Apparatus and method for speculatively executing instructions in a computer system|
|US5428807 *||Jun 17, 1993||Jun 27, 1995||Digital Equipment Corporation||Method and apparatus for propagating exception conditions of a computer system|
|US5446849 *||Nov 29, 1991||Aug 29, 1995||Kabushiki Kaisha Toshiba||Electronic computer which executes squash branching|
|US5454090 *||Apr 7, 1994||Sep 26, 1995||Siemens Aktiengesellschaft||Apparatus for furnishing instructions in a microprocessor with a multi-stage pipeline processing unit for processing instruction phase and having a memory and at least three additional memory units|
|US5481684 *||Jul 20, 1994||Jan 2, 1996||Exponential Technology, Inc.||Emulating operating system calls in an alternate instruction set using a modified code segment descriptor|
|US5481693 *||Jul 20, 1994||Jan 2, 1996||Exponential Technology, Inc.||Shared register architecture for a dual-instruction-set CPU|
|US5485589 *||Jul 22, 1994||Jan 16, 1996||Dell Usa, L.P.||Predictive addressing architecture|
|US5490255 *||Dec 26, 1991||Feb 6, 1996||Amdahl Corporation||Expedited execution of pipelined command having self-ordering operand processing requirements|
|US5490265 *||Apr 14, 1993||Feb 6, 1996||Intel Corporation||Late cancel method and apparatus for a high performance microprocessor system|
|US5511175 *||Oct 20, 1994||Apr 23, 1996||Nexgen, Inc.||Method an apparatus for store-into-instruction-stream detection and maintaining branch prediction cache consistency|
|US5542059 *||Dec 21, 1994||Jul 30, 1996||Exponential Technology, Inc.||Dual instruction set processor having a pipeline with a pipestage functional unit that is relocatable in time and sequence order|
|US5560036 *||May 1, 1995||Sep 24, 1996||Mitsubishi Denki Kabushiki Kaisha||Data processing having incircuit emulation function|
|US5574872 *||Jun 23, 1994||Nov 12, 1996||Intel Corporation||Method and apparatus for controlling the saving of pipelines in pipelined processors during trap handling|
|US5574927 *||Mar 25, 1994||Nov 12, 1996||International Meta Systems, Inc.||RISC architecture computer configured for emulation of the instruction set of a target computer|
|US5577259 *||Aug 9, 1994||Nov 19, 1996||Unisys Corporation||Instruction processor control system using separate hardware and microcode control signals to control the pipelined execution of multiple classes of machine instructions|
|US5611062 *||Mar 31, 1995||Mar 11, 1997||International Business Machines Corporation||Specialized millicode instruction for string operations|
|US5621909 *||Mar 12, 1996||Apr 15, 1997||International Business Machines Corporation||Specialized millicode instruction for range checking|
|US5623614 *||Sep 17, 1993||Apr 22, 1997||Advanced Micro Devices, Inc.||Branch prediction cache with multiple entries for returns having multiple callers|
|US5634136 *||Dec 8, 1995||May 27, 1997||Fujitsu Limited||Data processor and method of controlling the same|
|US5644761 *||Jun 5, 1992||Jul 1, 1997||Bull Hn Information Systems Inc.||Basic operations synchronization and local mode controller in a VLSI central processor|
|US5666507 *||Dec 26, 1995||Sep 9, 1997||Unisys Corporation||Pipelined microinstruction apparatus and methods with branch prediction and speculative state changing|
|US5680598 *||Mar 31, 1995||Oct 21, 1997||International Business Machines Corporation||Millicode extended memory addressing using operand access control register to control extended address concatenation|
|US5685009 *||Nov 29, 1995||Nov 4, 1997||Exponential Technology, Inc.||Shared floating-point registers and register port-pairing in a dual-architecture CPU|
|US5692170 *||Apr 28, 1995||Nov 25, 1997||Metaflow Technologies, Inc.||Apparatus for detecting and executing traps in a superscalar processor|
|US5694587 *||Mar 31, 1995||Dec 2, 1997||International Business Machines Corporation||Specialized millicode instructions for test PSW validity, load with access test, and character translation assist|
|US5748932 *||Jan 25, 1995||May 5, 1998||Advanced Micro Devices, Inc.||Cache memory system for dynamically altering single cache memory line as either branch target entry or prefetch instruction queue based upon instruction sequence|
|US5748976 *||Oct 18, 1993||May 5, 1998||Amdahl Corporation||Mechanism for maintaining data coherency in a branch history instruction cache|
|US5754810 *||Mar 12, 1996||May 19, 1998||International Business Machines Corporation||Specialized millicode instruction for certain decimal operations|
|US5754825 *||May 19, 1995||May 19, 1998||Compaq Computer Corporation||Lower address line prediction and substitution|
|US5758142 *||May 31, 1994||May 26, 1998||Digital Equipment Corporation||Trainable apparatus for predicting instruction outcomes in pipelined processors|
|US5761466 *||May 9, 1994||Jun 2, 1998||Lsi Logic Corporation||Soft programmable single-cycle/pipelined micro-programmed control system|
|US5761720 *||Mar 15, 1996||Jun 2, 1998||Rendition, Inc.||Pixel engine pipeline processor data caching mechanism|
|US5774357 *||Jun 6, 1995||Jun 30, 1998||Hoffberg; Steven M.||Human factored interface incorporating adaptive pattern recognition based controller apparatus|
|US5781750 *||Jan 11, 1994||Jul 14, 1998||Exponential Technology, Inc.||Dual-instruction-set architecture CPU with hidden software emulation mode|
|US5790821 *||Mar 8, 1996||Aug 4, 1998||Advanced Micro Devices, Inc.||Control bit vector storage for storing control vectors corresponding to instruction operations in a microprocessor|
|US5790844 *||Mar 31, 1997||Aug 4, 1998||International Business Machines Corporation||Millicode load and test access instruction that blocks interrupts in response to access exceptions|
|US5819078 *||Jun 10, 1997||Oct 6, 1998||International Business Machines Corporation||Addressing extended memory using millicode by concatenating a small millicode address and address extension data|
|US5822560 *||May 23, 1996||Oct 13, 1998||Advanced Micro Devices, Inc.||Apparatus for efficient instruction execution via variable issue and variable control vectors per issue|
|US5835949 *||Apr 2, 1997||Nov 10, 1998||National Semiconductor Corporation||Method of identifying and self-modifying code|
|US5860093 *||Jan 21, 1997||Jan 12, 1999||Unisys Corporation||Reduced instruction processor/storage controller interface|
|US5867699 *||Jul 25, 1996||Feb 2, 1999||Unisys Corporation||Instruction flow control for an instruction processor|
|US5875201 *||Dec 30, 1996||Feb 23, 1999||Unisys Corporation||Second level cache having instruction cache parity error control|
|US5878254 *||Apr 28, 1998||Mar 2, 1999||Hitachi, Ltd.||Instruction branching method and a processor|
|US5884057 *||May 8, 1995||Mar 16, 1999||Exponential Technology, Inc.||Temporal re-alignment of a floating point pipeline to an integer pipeline for emulation of a load-operate architecture on a load/store processor|
|US5893930 *||Jul 12, 1996||Apr 13, 1999||International Business Machines Corporation||Predictive translation of a data address utilizing sets of associative entries stored consecutively in a translation lookaside buffer|
|US5898850 *||Mar 31, 1997||Apr 27, 1999||International Business Machines Corporation||Method and system for executing a non-native mode-sensitive instruction within a computer system|
|US5928357 *||Jun 19, 1997||Jul 27, 1999||Intel Corporation||Circuitry and method for performing branching without pipeline delay|
|US5949995 *||Aug 2, 1996||Sep 7, 1999||Freeman; Jackie Andrew||Programmable branch prediction system and method for inserting prediction operation which is independent of execution of program code|
|US5954814 *||Dec 19, 1997||Sep 21, 1999||Intel Corporation||System for using a branch prediction unit to achieve serialization by forcing a branch misprediction to flush a pipeline|
|US5960455 *||Dec 30, 1996||Sep 28, 1999||Unisys Corporation||Scalable cross bar type storage controller|
|US5974543 *||Feb 17, 1998||Oct 26, 1999||International Business Machines Corporation||Apparatus and method for performing subroutine call and return operations|
|US6012141 *||Jun 19, 1997||Jan 4, 2000||Hyundai Electronics America||Apparatus for detecting and executing traps in a superscalar processor|
|US6088793 *||Dec 30, 1996||Jul 11, 2000||Intel Corporation||Method and apparatus for branch execution on a multiple-instruction-set-architecture microprocessor|
|US6092154 *||Nov 7, 1997||Jul 18, 2000||Intel Corporation||Method of pre-caching or pre-fetching data utilizing thread lists and multimedia editing systems using such pre-caching|
|US6157987 *||Nov 24, 1997||Dec 5, 2000||Micron Technology, Inc.||Pixel engine data caching mechanism|
|US6157988 *||Dec 23, 1997||Dec 5, 2000||Micron Technology, Inc.||Method and apparatus for high performance branching in pipelined microsystems|
|US6157994 *||Jul 8, 1998||Dec 5, 2000||Advanced Micro Devices, Inc.||Microprocessor employing and method of using a control bit vector storage for instruction execution|
|US6167509 *||May 16, 1994||Dec 26, 2000||Compaq Computer Corporation||Branch performance in high speed processor|
|US6185670 *||Oct 12, 1998||Feb 6, 2001||Intel Corporation||System for reducing number of opcodes required in a processor using an instruction format including operation class code and operation selector code fields|
|US6230260||Sep 1, 1998||May 8, 2001||International Business Machines Corporation||Circuit arrangement and method of speculative instruction execution utilizing instruction history caching|
|US6279098||Dec 16, 1996||Aug 21, 2001||Unisys Corporation||Method of and apparatus for serial dynamic system partitioning|
|US6289444||Jun 2, 1999||Sep 11, 2001||International Business Machines Corporation||Method and apparatus for subroutine call-return prediction|
|US6304962||Jun 2, 1999||Oct 16, 2001||International Business Machines Corporation||Method and apparatus for prefetching superblocks in a computer processing system|
|US6332215||Dec 8, 1998||Dec 18, 2001||Nazomi Communications, Inc.||Java virtual machine hardware for RISC and CISC processors|
|US6351804||Oct 10, 2000||Feb 26, 2002||Advanced Micro Devices, Inc.||Control bit vector storage for a microprocessor|
|US6378061 *||Nov 12, 1993||Apr 23, 2002||Intel Corporation||Apparatus for issuing instructions and reissuing a previous instructions by recirculating using the delay circuit|
|US6400996||Feb 1, 1999||Jun 4, 2002||Steven M. Hoffberg||Adaptive pattern recognition based control system and method|
|US6418424||May 4, 1999||Jul 9, 2002||Steven M. Hoffberg||Ergonomic man-machine interface incorporating adaptive pattern recognition based control system|
|US6438627 *||May 12, 1998||Aug 20, 2002||Compaq Information Technologies Group, L.P.||Lower address line prediction and substitution|
|US6543002 *||Nov 4, 1999||Apr 1, 2003||International Business Machines Corporation||Recovery from hang condition in a microprocessor|
|US6543003 *||Nov 8, 1999||Apr 1, 2003||International Business Machines Corporation||Method and apparatus for multi-stage hang recovery in an out-of-order microprocessor|
|US6604188 *||Oct 20, 1999||Aug 5, 2003||Transmeta Corporation||Pipeline replay support for multi-cycle operations wherein all VLIW instructions are flushed upon detection of a multi-cycle atom operation in a VLIW instruction|
|US6640145||Jun 3, 2002||Oct 28, 2003||Steven Hoffberg||Media recording device with packet data interface|
|US6647467||Oct 9, 2000||Nov 11, 2003||Micron Technology, Inc.||Method and apparatus for high performance branching in pipelined microsystems|
|US6678819 *||May 3, 2000||Jan 13, 2004||Inside Technologies||Pipeline microprocessor with conditional jump in one clock cycle|
|US6704861||Nov 19, 1996||Mar 9, 2004||Hewlett-Packard Development Company, L.P.||Mechanism for executing computer instructions in parallel|
|US6728865||Oct 20, 1999||Apr 27, 2004||Transmeta Corporation||Pipeline replay support for unaligned memory operations|
|US6807606||Dec 18, 2000||Oct 19, 2004||International Business Machines Corp.||Distributed execution coordination for web caching with dynamic content|
|US6810474 *||Mar 4, 1998||Oct 26, 2004||Hitachi, Ltd.||Information processor|
|US6823360||Dec 18, 2000||Nov 23, 2004||International Business Machines Corp.||Cofetching in a command cache|
|US6823447 *||Mar 1, 2001||Nov 23, 2004||International Business Machines Corporation||Software hint to improve the branch target prediction accuracy|
|US6826749||May 25, 2001||Nov 30, 2004||Nazomi Communications, Inc.||Java hardware accelerator using thread manager|
|US6877025||Dec 18, 2000||Apr 5, 2005||International Business Machines Corp.||Integrated JSP and command cache for web applications with dynamic content|
|US6904517||Nov 2, 2001||Jun 7, 2005||Arm Limited||Data processing apparatus and method for saving return state|
|US6907515||May 22, 2002||Jun 14, 2005||Arm Limited||Configuration control within data processing systems|
|US6965984||Apr 30, 2002||Nov 15, 2005||Arm Limited||Data processing using multiple instruction sets|
|US7000094||Jun 25, 2001||Feb 14, 2006||Arm Limited||Storing stack operands in registers|
|US7003652||Jun 25, 2001||Feb 21, 2006||Arm Limited||Restarting translated instructions|
|US7013382 *||Nov 2, 2001||Mar 14, 2006||Lsi Logic Corporation||Mechanism and method for reducing pipeline stalls between nested calls and digital signal processor incorporating the same|
|US7076771||Dec 1, 2000||Jul 11, 2006||Arm Limited||Instruction interpretation within a data processing system|
|US7080362||Aug 24, 2001||Jul 18, 2006||Nazomi Communication, Inc.||Java virtual machine hardware for RISC and CISC processors|
|US7089539||Feb 25, 2002||Aug 8, 2006||Arm Limited||Program instruction interpretation|
|US7131118||Jul 25, 2002||Oct 31, 2006||Arm Limited||Write-through caching a JAVAŽ local variable within a register of a register bank|
|US7134119||Jun 25, 2001||Nov 7, 2006||Arm Limited||Intercalling between native and non-native instruction sets|
|US7162611||May 2, 2002||Jan 9, 2007||Arm Limited||Unhandled operation handling in multiple instruction set systems|
|US7162621 *||Feb 21, 2001||Jan 9, 2007||Mips Technologies, Inc.||Virtual instruction expansion based on template and parameter selector information specifying sign-extension or concentration|
|US7197603||Sep 25, 2003||Mar 27, 2007||Micron Technology, Inc.||Method and apparatus for high performance branching in pipelined microsystems|
|US7225436||Oct 13, 2000||May 29, 2007||Nazomi Communications Inc.||Java hardware accelerator using microcode engine|
|US7231511 *||Dec 20, 2001||Jun 12, 2007||Intel Corporation||Microinstruction pointer stack including speculative pointers for out-of-order execution|
|US7243350 *||Sep 27, 2002||Jul 10, 2007||Koninklijke Philips Electronics N.V.||Speculative execution for java hardware accelerator|
|US7340733||Dec 5, 2002||Mar 4, 2008||International Business Machines Corporation||Optimizing source code for iterative execution|
|US7398358||Mar 27, 2007||Jul 8, 2008||Micron Technology, Inc.||Method and apparatus for high performance branching in pipelined microsystems|
|US7599981||Feb 21, 2001||Oct 6, 2009||Mips Technologies, Inc.||Binary polynomial multiplier|
|US7617388||Dec 22, 2006||Nov 10, 2009||Mips Technologies, Inc.||Virtual instruction expansion using parameter selector defining logic operation on parameters for template opcode substitution|
|US7636837 *||Apr 26, 2005||Dec 22, 2009||Fujitsu Limited||Apparatus and method for controlling instructions at time of failure of branch prediction|
|US7647489||Mar 14, 2005||Jan 12, 2010||Arm Limited||Function calling mechanism with embedded index for a handler program and an embedded immediate value for passing a parameter|
|US7673190 *||Sep 14, 2005||Mar 2, 2010||Unisys Corporation||System and method for detecting and recovering from errors in an instruction stream of an electronic data processing system|
|US7685403 *||Jun 16, 2003||Mar 23, 2010||Brett Coon||Pipeline replay support for multi-cycle operations|
|US7702800||Dec 18, 2000||Apr 20, 2010||International Business Machines Corporation||Detecting and handling affinity breaks in web applications|
|US7711763||Feb 21, 2001||May 4, 2010||Mips Technologies, Inc.||Microprocessor instructions for performing polynomial arithmetic operations|
|US7802080||Mar 24, 2004||Sep 21, 2010||Arm Limited||Null exception handling|
|US7860911||Apr 25, 2006||Dec 28, 2010||Mips Technologies, Inc.||Extended precision accumulator|
|US7886135||Nov 7, 2006||Feb 8, 2011||Brett Coon||Pipeline replay support for unaligned memory operations|
|US7930526||Mar 24, 2004||Apr 19, 2011||Arm Limited||Compare and branch mechanism|
|US7974714||Aug 29, 2006||Jul 5, 2011||Steven Mark Hoffberg||Intelligent electronic appliance system and method|
|US8046313||Nov 13, 2006||Oct 25, 2011||Hoffberg Steven M||Ergonomic man-machine interface incorporating adaptive pattern recognition based control system|
|US8117423||Mar 4, 2008||Feb 14, 2012||Brett Coon||Pipeline replay support for multicycle operations|
|US8176286||Feb 20, 2004||May 8, 2012||Edward Colles Nevill||Memory recycling in computer systems|
|US8185882||Feb 14, 2006||May 22, 2012||Nazomi Communications Inc.||Java virtual machine hardware for RISC and CISC processors|
|US8369967||Mar 7, 2011||Feb 5, 2013||Hoffberg Steven M||Alarm system controller and a method for controlling an alarm system|
|US8412824 *||Aug 27, 2009||Apr 2, 2013||Symantec Corporation||Systems and methods for dynamically managing the migration of a single instance of data between storage devices|
|US8447958||Mar 6, 2009||May 21, 2013||Bridge Crossing, Llc||Substituting portion of template instruction parameter with selected virtual instruction parameter|
|US8473718||Oct 3, 2006||Jun 25, 2013||Nazomi Communications Inc.||Java hardware accelerator using microcode engine|
|US8516224||Jan 31, 2012||Aug 20, 2013||Brett Coon||Pipeline replay support for multicycle operations|
|US8578347 *||Dec 28, 2006||Nov 5, 2013||The Mathworks, Inc.||Determining stack usage of generated code from a model|
|US8583263||Mar 8, 2011||Nov 12, 2013||Steven M. Hoffberg||Internet appliance system and method|
|US8769508||Jun 29, 2005||Jul 1, 2014||Nazomi Communications Inc.||Virtual machine hardware for RISC and CISC processors|
|US8892495||Jan 8, 2013||Nov 18, 2014||Blanding Hovenweep, Llc||Adaptive pattern recognition based controller apparatus and method and human-interface therefore|
|US20020069402 *||Dec 7, 2000||Jun 6, 2002||Nevill Edward Colles||Scheduling control within a system having mixed hardware and software based instruction execution|
|US20020108103 *||Jun 25, 2001||Aug 8, 2002||Nevill Edward Colles||Intercalling between native and non-native instruction sets|
|US20020111992 *||Dec 18, 2000||Aug 15, 2002||Copeland George P.||JSP composition in a cache for web applications with dynamic content|
|US20020116448 *||Dec 18, 2000||Aug 22, 2002||Copeland George P.||Cofetching in a command cache|
|US20020116603 *||Feb 21, 2001||Aug 22, 2002||Ownership||Configurable instruction sequence generation|
|US20020157000 *||Mar 1, 2001||Oct 24, 2002||International Business Machines Corporation||Software hint to improve the branch target prediction accuracy|
|US20020188825 *||Apr 30, 2002||Dec 12, 2002||Seal David James||Data processing using multiple instruction sets|
|US20020188826 *||May 2, 2002||Dec 12, 2002||Rose Andrew Christopher||Unhandled operation handling in multiple instruction set systems|
|US20030084271 *||Sep 27, 2002||May 1, 2003||Lindwer Menno Menasshe||Speculative execution for java hardware accelerator|
|US20030115579 *||Dec 5, 2002||Jun 19, 2003||International Business Machines Corporation||Optimizing source code for iterative execution|
|US20030191929 *||Dec 20, 2001||Oct 9, 2003||Cornaby Michael P.||Microinstruction pointer stack in a processor|
|US20040015896 *||Feb 25, 2002||Jan 22, 2004||Dornan Christopher Bentley||Program instruction interpretation|
|US20040019880 *||Jul 25, 2002||Jan 29, 2004||Rose Andrew Christopher||Write-through caching a java local variable within a register of a register bank|
|US20040068643 *||Sep 25, 2003||Apr 8, 2004||Dowling Eric M.||Method and apparatus for high performance branching in pipelined microsystems|
|US20040073630 *||Dec 18, 2000||Apr 15, 2004||Copeland George P.||Integrated JSP and command cache for web applications with dynamic content|
|US20040193828 *||Feb 20, 2004||Sep 30, 2004||Arm Limited||Memory recycling in computer systems|
|US20050149694 *||Feb 17, 2005||Jul 7, 2005||Mukesh Patel||Java hardware accelerator using microcode engine|
|US20050188187 *||Apr 26, 2005||Aug 25, 2005||Fujitsu Limited||Apparatus and method for controlling instructions at time of failure of branch prediction|
|US20050210226 *||Mar 14, 2005||Sep 22, 2005||Arm Limited||Function calling mechanism|
|US20060095752 *||Aug 28, 2004||May 4, 2006||Min-Cheng Kao||[method for return instruction identification and associated method for return target pointer prediction]|
|US20070106889 *||Dec 22, 2006||May 10, 2007||Mips Technologies, Inc.||Configurable instruction sequence generation|
|US20070118724 *||Oct 3, 2006||May 24, 2007||Nazomi Communications Inc.||Java hardware accelerator using microcode engine|
|US20070174599 *||Mar 27, 2007||Jul 26, 2007||Micron Technology, Inc.||Method and apparatus for high performance branching in pipelined microsystems|
|US20080028381 *||Oct 10, 2007||Jan 31, 2008||International Business Machines Corporation||Optimizing source code for iterative execution|
|US20100064118 *||Sep 10, 2008||Mar 11, 2010||Vns Portfolio Llc||Method and Apparatus for Reducing Latency Associated with Executing Multiple Instruction Groups|
|USRE45278 *||Oct 28, 2010||Dec 2, 2014||Stmicroelectronics, Inc.||Method and apparatus for changing microcode to be executed in a processor|
|WO1993017385A1 *||Feb 23, 1993||Sep 2, 1993||Intel Corporation||Dynamic flow instruction cache memory|
|U.S. Classification||712/240, 712/E09.083, 712/E09.057|
|International Classification||G06F9/38, G06F9/42|
|Cooperative Classification||G06F9/3806, G06F9/4426|
|European Classification||G06F9/38B2B, G06F9/44F1A1|
|Jun 27, 1988||AS||Assignment|
Owner name: PRIME COMPUTER, INC., PRIME PARK, NATICK, MA 01760
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST.;ASSIGNORS:BECKWITH, ROBERT F.;JOHNSON, NEIL J.;IRUKULLA, SUREN;AND OTHERS;REEL/FRAME:004967/0311
Effective date: 19880623
Owner name: PRIME COMPUTER, INC., MASSACHUSETTS
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:BECKWITH, ROBERT F.;JOHNSON, NEIL J.;IRUKULLA, SUREN;ANDOTHERS;REEL/FRAME:004967/0311
Effective date: 19880623
|Mar 16, 1990||AS||Assignment|
Owner name: CHEMICAL BANK (A NEW YORK BANKING CORPORATION), NE
Free format text: SECURITY INTEREST;ASSIGNORS:DR HOLDINGS INC., A DE CORP.;DR ACQUISITION CORP., A CORP. OF DE;PRIME COMPUTER INC.;AND OTHERS;REEL/FRAME:005333/0131
Effective date: 19900130
|Nov 2, 1992||AS||Assignment|
Owner name: CHEMICAL BANK, A NY CORP., NEW YORK
Free format text: SECURITY INTEREST;ASSIGNOR:COMPUTERVISION CORPORATION, A CORP. OF DE;REEL/FRAME:006314/0077
Effective date: 19920821
|Jun 1, 1993||AS||Assignment|
Owner name: COMPUTERVISION CORPORATION, MASSACHUSETTS
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:PRIME COMPUTER, INC.;REEL/FRAME:006663/0565
Effective date: 19920813
|Mar 12, 1996||REMI||Maintenance fee reminder mailed|
|Apr 17, 1996||SULP||Surcharge for late payment|
|Apr 17, 1996||FPAY||Fee payment|
Year of fee payment: 4
|Dec 1, 1997||AS||Assignment|
Owner name: CVSI, INC., MASSACHUSETTS
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:COMPUTERVISION CORPORATION;REEL/FRAME:008829/0423
Effective date: 19970926
Owner name: FARRINGTON INVESTMENTS, LTD., CAYMAN ISLANDS
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:CVSI, INC.;REEL/FRAME:008829/0427
Effective date: 19970926
|Apr 20, 1998||AS||Assignment|
Owner name: CHASE MANHATTAN BANK (F/K/A CHEMICAL BANK), AS COL
Free format text: TERMINATION AND RELEASE OF ASSIGNMENT OF SECURITY INTEREST IN PATENTS;ASSIGNOR:COMPUTERVISION CORPORATION, A DELAWARE CORPORATION;REEL/FRAME:009178/0329
Effective date: 19980417
|Jun 11, 1998||AS||Assignment|
Owner name: FARRINGTON INVESTMENTS LTD., CAYMAN ISLANDS
Free format text: RE-RECORD TO CORRECT ASSIGNEE ADDRESS PREVIOUSLY RECORDED REEL/FRAME 8829/0427 .;ASSIGNOR:CVSI, INC.;REEL/FRAME:009463/0527
Effective date: 19970926
|Jan 3, 2000||AS||Assignment|
Owner name: INTEL CORPORATION, CALIFORNIA
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:FARRINGTON INVESTMENTS, LTD.;REEL/FRAME:010485/0638
Effective date: 19990913
|Feb 3, 2000||FPAY||Fee payment|
Year of fee payment: 8
|Feb 4, 2004||FPAY||Fee payment|
Year of fee payment: 12
|Feb 18, 2004||REMI||Maintenance fee reminder mailed|